prophys.explain

Explanations of surrogates and of whole probabilistic models.

Local methods attribute single predictions of a surrogate to its inputs. System methods attribute a risk statistic of a model attribute to the model’s random inputs, including the epistemic uncertainty of its surrogates.

Classes

ShapValues	Feature attributions of individual predictions.
SystemShapley	Shapley attribution of a statistic of one attribute to its random inputs.

Functions

as_function	Wrap a surrogate, tree ensemble or callable as a batch prediction function.
expected_gradients	Expected gradients attributions of a differentiable model.
integrated_gradients	Integrated gradients attributions of a differentiable model.
kernel_shap	Approximate interventional Shapley values by KernelSHAP.
shapley_effects	Attribute a risk statistic of a model attribute to its random inputs by Shapley values.
shapley_values	Exact interventional Shapley values and pairwise interactions, for up to 16 features.
tree_shap	Exact path dependent TreeSHAP attributions of a tree ensemble.